Banter or Threat?
Humans beat the filter on gamer trash talk, and appeals become fair across languages.
The job
Finds where automatic moderation mistakes gamer banter in other languages for threats, and misses real threats. Fixes the rules, reinstates wrongly penalised players, and proves appeals are equally fair for English and non-English players.
The moment
Every day, human moderators and the automatic filter take the same quiz on gamer slang: banter or threat?
Today's result: moderators 18 out of 20, the automatic filter 7 out of 20.
New rules ship. Non-English players used to win appeals 3.0 times as often as English players, a sign they were more often wrongly penalised. Now it is 1.1 times.
400 wrongly banned players get their accounts back with an apology in their own language, and the same review found 12 real threats the old filter missed.
What it does
- Record an appeal verdict
- Submit quiz answers
- Send allow/block patterns to the moderation vendor
- Reinstate and apologise
- Escalate a real threat to a human
What you see
Courtroom appeal docket with a daily head-to-head 'banter or threat?' quiz (humans vs the automod) and a language parity gauge
What it moves
How often non-English players win appeals compared with English players, wrongly banned players reinstated, and real threats caught each month.
Built for
- Risk & fraud
- Marketing, brand & comms
- Contact-centre operations
